SONM Platform — a half year checkpoint

9 min readFeb 18, 2019


New Year has finally come both in Gregorian and Chinese calendars so this is a good moment to take a look back, to sum up the work done.

At the moment, the whole crypto world is in depression, some projects have already performed their «coming-out» and it seems that we will witness very entertaining news this year. SONM, despite these turbulent times, continues to work hard on building a platform, that was launched in beta in July 2018.

After half a year we want not only to share the platform’s early traction but to explain why each component was developed, since it’s very important for understanding our future strategy.

Platform components

At the end of July 2018, we launched SONM beta to test how all the key components work and collect feedback from the first customers:

1. Sidechain

We’ve launched own Ethereum sidechain for internal transactions and smart contract execution.

As we all know, Ethereum is capable to perform near ~20 transactions per second, and the introduction of new solutions that would help Ethereum to scale is constantly postponed.

More importantly, for renting computing power, our business logic requires a significant number of transactions. On the Ethereum mainnet, the cost of these transactions would exceed the cost of renting the computing resources itself.

This is not a problem of only the Ethereum network, but any platform will also have almost the same limitations, mainly due to the classical CAP-theorem. Our words are not a criticism of Ethereum, but simply a history of dealing with difficulties.

Switching to own sidechain enabled SONM to make transactions extremely cheap and fast today, as well as to independently work on scalability options for the future.

In the coming years, when SONM will serve thousands of customers and perform millions of computations, we will need to process hundreds of thousands or even millions of transactions per second.

With a sidechain SONM will be able to implement sharding (on application level) or, as a last resort, to gradually switch to own network, designed to manage distributed infrastructure and perform useful computations at scale.

We haven’t made a final decision and continue to carefully study all the options. As soon as the optimal solution for the network speed and scalability is found, we will definitely publish an announcement. But, at least in early 2019, the network is likely to remain as it is.

2. Marketplace

We’ve introduced a Marketplace, that works much like other spot markets: GPU rigs and enterprise servers owners can connect their devices to the platform and get paid for performing useful computations. This is very similar to the Uber drivers app, which allows both private drivers and taxi companies to join the platform.

Moreover, we plan to introduce a rating for each supplier, similar to the rating of drivers in Uber. This rating will be based on whether the device is 24/7 online, what percent of the tasks the supplier has completed without fails, hardware characteristics, network bandwidth, etc. According to this rating, the Resource Allocation System will automatically allocate a specific task to the most suitable nodes.

The second important goal we are focused on is to simplify the connection of any device to the platform. We are sure that this should happen in just a couple of clicks, something like creating an account on a new website or in a mobile application. In a few years, we plan to get more GPU instances than large cloud providers have.

According to our estimates, AWS currently has about 1.8 million Servers, but only about 10% are optimized for GPU-related computations.

SONM starts with the numbers that Amazon had in 2009, but thanks to our Shared Economy model, we can connect new devices much faster, as we don’t need to build data centers, and we don’t have capital expenses for buying and maintaining servers.

To become one of the leaders in the number of GPU instances, we will need to connect about 50–100k cards to the platform. You can find more details in the “Supply traction” section below.

3. Resource Allocation System

Think of Marketplace as our own infrastructure (hardware layer), and the Resource Allocation System as a set of tools (middleware) for intelligently managing the hardware collected in this data center.

In other words, RAS is an early version of the future infrastructure management & orchestration system. At the moment, RAS includes a set of components that automatically perform all actions to rent the right amount of instances at the Marketplace and run computations.

This component creates a common mapping of resources on the network according to current availability, hardware characteristics (GPU, CPU, RAM, Storage) and computing hour price set by the supplier.

Task Management
Based on data from the Mapping component, Task Manager places an order for the rental of required resources, concludes deals with suppliers and performs computations.

Provides clients with logs and dashboards for all stages of the task execution.


We developed an operating system distribution designed to help suppliers deploy a SONM worker on their device in a couple of minutes.

This is an early version of what will become a trusted OS, designed for securing the computations on consumer devices and preventing the device owner from decrypting and copying customer data.

We will share more detailed docs about our plans to build a Trusted Execution Environment in coming weeks.

All of the above components, in essence, form a distributed Data Center based on a Shared Economy model (i.e. consisting of consumer devices), as well as intelligent middleware that manages bare devices in this data center.

Further platform development is focused on creating the world’s largest Data Center powered by an increasing number of suppliers around the globe, and democratizing the Cloud Computing market by reducing the price of computing hour up to 10 times.

Supply traction:

Over the past six months, more than 5,000 GPUs and 6,000 Enterprise Servers were connected to the SONM platform. It all happened organically since we didn’t push supply until this point. Today, the distribution of workers covers 34 countries (and continues to grow):

USA, Canada, India, South Korea, Singapore, Australia, Japan, China, Germany, Ireland, UK, Sweden, France, Brazil, Egypt, Russia, Poland, Slovakia, Spain, Austria, Slovenia, Ukraine, Lithuania, Hungary, Bulgaria, Greece, Italy, Netherlands, Argentina, Vietnam, New Zealand, Malaysia, South Africa, Norway

SONM is already present in more countries than AWS and Azure, since they only have data centers in 14 and 11 countries, respectively.

In the coming year, SONM will focus on expanding the number of countries of presence and increasing the number of suppliers.

We see 3 types of devices that can potentially be connected to the platform:

As a first goal, SONM is going to attract 2,000,000 GPUs owned by cryptocurrency miners. Mining has become unprofitable and most of the miners are looking for other ways to utilize their GPU rigs. They will be able to switch their devices to SONM platform and get paid for performing useful computations. Potentially, their income will be comparable to the mining or even higher.

Demand traction:

Since our main goal was technology testing, we didn’t focus on mass adoption, aggressive sales and generating profits. Our main goal was to attract a few companies from the industries that need GPU-related computations and collect feedback.

Over the past six months, we have tested the platform on the following use cases:

1. Video Rendering

The Shotty post-production studio has acquired the resources needed for video rendering, allowing a significant reduction in production costs with a saving of up to 80%.

Another use case was renting additional computing resource by PostKino FX studio to cut the rendering time and meet the deadline.

2. CDN

Teleport CDN rents computing power, disk, and network resources from SONM in the moments when streaming projects and peak loads require additional resources. Teleport has its own network of devices, but it uses SONM as an on-demand source of computing power.

3. Machine Learning

Ashmanov Neural Networks develops machine learning systems for data analysis. With our help, the company was able to acquire the resources it needed for training deep neural networks. Pilot testing showed that they can rent computing power 5 times cheaper in comparison to the cost of the cloud providers.

4. Telegram proxy servers

In 2018, Telegram was blocked in Russia and some other countries due to the refusal to provide authorities with access to the users’ message history. In fact, Telegram entered the fight with the authorities, who constantly blocked all known Telegram IP addresses, and Telegram constantly raised new proxy servers. After a few months of stubborn battles, Telegram won and continues to work successfully on the territory of these countries. SONM has also become a part of this story — you surely remember this story under the code name “Operation A”.

Working with these companies we got a lot of important insights about their real needs, about the challenges they face when using traditional Cloud, about the features they would like to get from SONM, and what barriers prevent them from adopting our services.

Demand traction for the second half of 2018:

In addition to the companies mentioned above, 112 unique customers have rented computing resources at SONM.

They placed over 1,000,000 orders to buy computing resources on the Marketplace.

And almost 140,000 of these orders turned into deals (deal = when computations were made and reward paid)

And more than 410,000 computing hours were performed on these deals:

These numbers seem encouraging for us, because such organic growth in demand is a pretty good result for a young startup that didn’t push any promotion, and released just an early beta of its product to test the market fit.

We are sure that after the product is completed and we start global promotion, SONM will be able to achieve impressive results and start generating first profit.

What the future holds?

Now, when the platform beta is launched and tested on the real world cases, the time has come for SONM to enter to a new development stage, to get down to adoption and product monetization to expand to the real world markets and draw new clients.

In the coming days, we will be happy to share our platform development plan for the next 3 years (as well as a business strategy), based on insights received from our first customers and on our understanding of the current market situation.




Global Fog Computing Platform